civitai is absolutely shitting its self with this upload.... not sure whats going on. going to wait until information and changes finally get through the clogged pipes of this site.
MERGE WARNING!
Model trained on deliberately falsely tagged data set.. Many styles given ... a bit over the top? But yeah to get the look of this I've used images from flux schnell, photographs of a spicy nature, and some boring ones too and various styles pulled out of SDXL and even a dash 1.5. All tagged with: real life (must be in prompt) and no other style tags. Tags are also incomplete, a work in progress as the data set grows. So think twice before using this in a merge.
But, if you do use it anyway, please respect the work done here and give me a shout out and link the model, and maybe, just maybe...
(Insert reasons of cost and time spent here)...then, (persuasion:666), If this model helps you with your AI, erm, explorations and you want to see where this is going,,, well, BUZZ, will help me, help you.
SOOOooo..... Now that I'm done warning and begging, what's going on?
Photorealism has been given the boot. Here at Dank Labs we've (me|myself|& I:1.1) decided that UncannyAF should live up to its name. Chapter 3 has been abandoned, spent waaaay too much time trying to salvage the broken Clip model that made it an almost un-promptable exercise in patients. But fear not. Chapter 2 stepped up, and, immediately regretted it as I shoved it in the oven to get gud. And after several Rocky montages we're here... on the first step of that bloody flight of stairs. There's a long way to go, but i feel it's worthy of a release.
To Tame the Beast:
CFG: 4 to 8
Step: 20 and beyond.
EulerA Simple
The preview images are made in Comfy so right click, save and drop em in for prompts and parameters and shite workflow.
For smaller images I fell adetailer/face detailer are a must, when you get into upscaling tho, each image is its own thing.
PROMPTING:
Always start with the usual: score_9,score_8_up,score_7_up,score_6_up,score_5_up,score_4_up
Always add: Real life.
And then, in the negative I tend to have: big eyes, 3d, cgi, sfm, anime style, digital art, render,,, but I'm not sure if it does anything tbh.
So,,, have at it.
Road Map:
Fuck it, guna keep it in green to steal your attention. Make it edgy with italics,,, damn. The direction is what you'd expect. Expand the data set and train it. In the next chapter I want to focus on improving (in order of importance):
Background diversity and fidelity. Biggest hole in the story so far imo.
Tagging. Orcs shouldn't be goblins and goblins shouldn't be orcs. Just ain't right.
Men. I hear they extend beyond the penis. Time to find out.
Faces and eyes. Same-face is what it is, you average enough 1girls and you land where you land. I want my same-face to be better.
Metals and materials. For metalers and materialers.
LEGACY INTRO BELOW.
So, what is this model... well, not to beat about the bush, it's a smut factory. I took pony v6 and tried pushing it towards the photographic. It's not quite there, but honestly I'm not sure I want it to be. This strikes a balance between realism and style that I find to be... erm... aesthetically pleasing.
It's a heavily Lora-ed model so how it plays with other Loras is anyone's guess. The process was: aggressively merging loras into the model in Supermerger. Merging the broken lora-ed model and the original using various block weight presets. This produced a collection of horrid little blighters that I alternatively threw into a collection of cement mixers, blenders and a toaster. One escaped up a chimney. More Loras got heaped on or smootched in. It was brutal.
After a week or so of activities that are probably in breach of the future "AI model integrity rights bill" UncannyAF splatted its way onto the pavement. You probably shouldn't download it. No good can come from it. Honestly. Not good for your mental health this one.
Anyway, you're still bloody here so... Being a Pony spawn you need to start your prompt with this piece of poetry:
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, {prompt}
Proof that, if you make something good enough, people will adopt anything. Anyway, copy and paste the above into a style and save for ease of use. Tokens I use pretty much for every generation are:
Analog style
Real life
Parameters:
cfg: 4 until 8. The higher you go the more 3d render the style gets.
Clipskip: 1 or 2 (I honestly haven't noticed the difference)
Steps: 42. Don't panic.
Hiresfix:
Denoise: 0.1 to 0.25. Sometimes you can go higher but mostly best to keep it low. Mostly. Damn it Newt, not now.
Steps: I like using 20 or so. Probably excessive.
Upscale: 2
Adetailer: The vast majority of the time I leave this alone, but if not, try it with only the eyes before going into the full face or anything more extreme.
Soooo. Now for the unfortunate ingredients. First off, huge thank you to the makers of all of these wonderful creations. And, sorry for what I did to em.
(The brood mama)
https://civarchive.com/models/257749/pony-diffusion-v6-xl
(The broodlings)
https://civarchive.com/models/366427/berserk-style-horror
https://civarchive.com/models/280472/chiaroscuro-contrasted-lighting-style-xl
https://civarchive.com/models/119376/better-faces-sdxl
https://civarchive.com/models/256740/perfect-eyes-sdxl
https://civarchive.com/models/135160/polyhedronall-sdxl-10-skin-hands-eyes-mf
https://civarchive.com/models/129689/analogredmond-analog-style-photography-lora-for-sd-xl
https://civarchive.com/models/259627/bad-quality-lora-or-sdxl
https://civarchive.com/models/122359/detail-tweaker-xl
https://civarchive.com/models/124347/xlmoreart-full-xlreal-enhancer
https://civarchive.com/models/235495/cinematic-kodak-motion-picture-film-still-style-xl
https://civarchive.com/models/214956/cinematic-photography-style-xl
I know. I know. There's no excuse for it.
Description
FAQ
Comments (30)
The pony is characterized by exaggeration that is unique to 2D, and the pony does a better job than other xl models, and I think it would be unwise to lose this feature if it were to be made 3D or realistic.
Bruh, let OP cook. I like the result here.
PonyXL still exists, so, no worries man.
Gunna keep this previous comment here for the sake of honesty:
(it's a fair point sir. I would however suggest that a self proclaimed smut factory is not the place to seek wisdom.)
Sorry, I literally woke up, read this, and responded. Not a morning guy, and there is for sure already enough snappy bad attitudes online. So again, sorry. Please allow my to answer in good faith rather the morning arse wipe.
While I agree in the assessment of what Pony is, I think it fair to point out that it wouldn't exist at all had the creator decided to respect SDXL and not try and make something different from it. Also that exaggerated flair PonyXL has is exactly the right kind of zest that I feel is missing from SDXL nsfw models. So hopefully, I can make something really good.
So. Mr Wang. I wish you a good day. And hope that you are unwise enough to enjoy the model.
I'm really sorry, I'm not saying that this model is bad, your attempt is worth encouraging, I just expressed my opinion on the difference between 3D and 2D, I apologize for my rashness, your attempt is also a process of improvement, I wish you all the best
Pony should be used to improve the responsiveness of other models and allow realistic models to benefit from its training. Segregating these things won't do any of us any favors and there are flavors for all of us. Most SDXL models are very poor at rendering much more than what the base SDXL model is capable of (for example - the same face on every creation).
Very few models are exceptional while models like what OP is doing are helping to push things forward out of the stagnant pool and into new frontiers.
Realistic doesn't have to be boring and anime doesn't have to over the top.
I appreciate models like this one because they're actually interesting and trying to push beyond boundaries.
Imagine being able to actually type "black woman" in a pony model and you get a black woman. Right now you can't do that. You have to use loras or beat it over the head until it understands that you don't want an asian chick with pale skin or the default SD model with dark skin.
Pony has limitations as do none-pony models. That doesn't have to be the case.
@LetTheBassDrop Agreed and Pony overwrote pretty much all of base SDXL in its training. So it can't do things like text well at all or use basic SDXL LoRAs well. The better we can merge to a middle ground, the happier we'll all be. 😝
@wang1818 Don't feel bad, happens to everyone. Sometimes we hit 'comment' without thinking /should/ we really comment. Everyone has opinions, but sometimes they are destructive rather than constructive. But no worries. We're just all here for the waifus. 😂
@wang1818 No sweat man, was not my finest response either. The thing with online talk is it lacks all context and nuance, lacks both the richness and guard rails of face to face banter. And in the end there seems to be a bit of decent discussion going on in here atm. So, the job's a good one.
@LetTheBassDrop I agree. The only real criticism I have of PonyXL is the choice to rewrite the clip model. It's fair enough, you put in the work (and fine tuning is way more work than merging), you get to set the terms of your effort. But man, I wish I could get mixing it with other SDXL models, god knows I tried but it's like mixing oil and water. Just no way... other than loras. The theory being it should be possible to drip feed some of SDXL's fidelity into the chaotic beauty of Pony. No idea if it'll work out in the long run, but so far so good I'd say. Searching for that middle ground @Jianka was talking about.
Very nice! Thank you! It makes our favorite characters look amazing! Downloading now! 💖
Jianka, thank you for the kind words. I really hope you enjoy exploring this beast as much as I did making it.
What an entertaining description :)
Here, in the Uncanny sewer, we give home to only the highest quality of pros. I'm happy it tickled ye. Have a good one.
This is the cleanest and aesthetically one of the better Pony mods made.
Thanks Op, this is wonderful work and a nice humorous write up.
Cheers!
I do declare, you sa, are making me blush. Seriously though really appreciate that you took to time to leave such a nice comment. Have yer self a good day! And I'm glad you're enjoying the model.
Great for making realistic refs to send to Elon Musk of how I want him to make a Waifur Teslabot for me. 😂
first time I've herd that mentioned as a use case. Considering what this thing is capable of he'll have his work cut out for him. :)
@DankGabrillo Only way he'll nail the look I want is with good realistic references. 😂
Yeah, this is my favorite Pony model so far. My goodness, it's capable of incredible visuals.
Welcome to the fold. Hope Uncanny can help in your task of liberation! Thanks for the kind words and review.
@DankGabrillo Oh, it most certainly will. I'll try and share more creations on your page to help people see what's possible. :)
@waifuliberator may the waifupocolypse begin!
Chapter 2 please
in the pipe, expect it this week end of not before :)
...granted.
Not a lot of what others are using it for, but by far the most realism I've managed to get for furs in a pony model. Just about any pony model can do 2D furs well, but none of them have done realistic as well as this. Generic was close, at points, but after settling in and really finding good prompts, Uncanny blows it away.
Woaw man, high praise. Really glad you're getting good use out of it. Would love to see some of your generation, could be good input for the next version.
@DankGabrillo I'm trying. I am still having to inpaint stuff and unfortunately for that one little section of 15 pixels I'm having to blow up the inpaint prompt or manually correct, but for 99% of the image, it hits. I have a few I'm trying to upload but the site doesn't want to cooperate, lol.
@DankGabrillo Posted! They may not be conducive to training in the dataset because of the dimensions (1920x800, 3840x1600) but still. I will work on more when I can, of more "normal" resolutions.
@Kaladae thats great! It's more to see your prompt style and what you're getting out of the model. A few words can make a huge difference when balancing a prompt with this model and fur wasn't something i was paying attention to. I'll pay more attention to it in Chapter 3. Chapter 2 will be landing tomorrow (hopefully) so, consider yourself the official fur tester.
@DankGabrillo Hey awesome. Make sure and be mindful of which specific images are actually attached to the checkpoint if you're looking at the post because I'm just using the post as a dump for all similar kinds of images. Unless you want to use them all, feel free lol.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.